Dynamic configuration

[v01t]

Hello,

As I understood models enabled with corresponding lines in elasticsearch.yml, like: ingest.opennlp.model.file.persons: en-ner-persons.bin and such configuration parameters loaded during start, so in order to extend configuration, we need to restart the node or there is a way to do so without restart currently?

Otherwise, is there any plans to have dynamic configuration and store such information (links to models) in a dedicated index so new models can be enabled without restart? (with config reload from the index every x min)

thanks for a great ingest processor!

[spinscale]

The files are in their own format and do not correspond to anything that could be indexed.

Reloading models can be achieved using the ResourceWatcherService (see https://github.com/elastic/elasticsearch/blob/master/server/src/main/java/org/elasticsearch/watcher/ResourceWatcherService.java)

If you want to work on that, I'll be happy to assist and review the PR.

[v01t]

It's not about files itself but their declaration in config

On Wed, Mar 7, 2018, 01:11 Alexander Reelsen [email protected] wrote:

The files are in their own format and do not correspond to anything that could be indexed.

Reloading models can be achieved using the ResourceWatcherService (see https://github.com/elastic/elasticsearch/blob/master/server/src/main/java/org/elasticsearch/watcher/ResourceWatcherService.java )

If you want to work on that, I'll be happy to assist and review the PR.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/spinscale/elasticsearch-ingest-opennlp/issues/23#issuecomment-370975640, or mute the thread https://github.com/notifications/unsubscribe-auth/ADGsjHIufKuad9lPMVXowXvcs9SDMVCuks5tbyW7gaJpZM4SfmVY .

[spinscale]

oh, then you need to make the MODEL_FILE_SETTINGS in the plugin dynamic and register a cluster state listener so you can do reloading there

[v01t]

Is there any chances to have this feature in near feature?

requirement to restart the cluster in order to apply changes/extend list of models files makes plugin incompatible with any production clusters that continiously receive realtime events and used for others purposes so restart in order to reconfigure just one plugin is unacceptable in that case.

Thanks for a great plugin!

[spinscale]

I am happy to accept pull requests and guide other people towards a well tested implementation, but I do not intend to work on this in the near future myself.